SRCB at the NTCIR-16 Real-MedNLP Task

Key Points

• The research aims to tackle Few-resource Named Entity Recognition and Adverse Drug Event detection challenges in medical NLP.
• Participated in NER and ADE subtasks at NTCIR-16 Real-MedNLP
• Developed NER systems using pretraining models, span-based classification, and prompt learning
• Employed data augmentation and model ensemble techniques
• Used multi-class classification and prompt learning for ADE detection
• Applied transfer learning and a two-stage training strategy to address long tail distribution issues
• Achieved improvements in few-resource NER performance through innovative systems
• Demonstrated effective ADE detection using multi-class classification
• Results highlight the benefits of ensemble methods and transfer learning in challenging NLP tasks